Doping effects on the structural and optical properties of GaN

(1)

Linköping Studies in Science and Technology

Dissertation No. 1554

Doping effects on the structural and optical

properties of GaN

Sergey Khromov

Thin Film Physics Division

Department of Physics, Chemistry, and Biology (IFM)

Linköping University

(2)

© Sergey Khromov

ISBN: 978-91-7519-483-7

ISSN 0345-7524

Printed by LiU-Tryck

Linköping, Sweden, 2013

(3)

A

BSTRACT

Today there is a strong drive towards higher efficiency light emitters and devices for power electronics based on GaN and its ternary compounds. Device performance can be improved in several ways on the material level. Development of bulk GaN to substitute sapphire and SiC as substrate materials can allow lower defect density epitaxial GaN layers to be grown. Using nonpolar homoepitaxial layers alleviates the problem of polarization fields present in polar GaN epilayers. This thesis advances the field by attacking outstanding problems related to doping and its influence on structural and optical properties of GaN. Optical and structural investigations were performed on bulk GaN grown by halide vapor phase epitaxy (HVPE) and on polar and nonpolar epitaxial GaN grown by metal organic chemical vapor deposition (MOCVD), doped with different impurities: Mg, Si, O or C. Optical characterization was done using photoluminescence (PL), time-resolved photoluminescence (TRPL), and cathodoluminescence (CL) in-situ scanning electron microscope, whereas structural properties were studied by means of transmission electron microscopy (TEM) and atom probe tomography (APT).

A correlation between Mg doping levels and stacking fault (SF) concentration in highly Mg-doped c-plane homoepitaxial GaN layers is found. Increasing Mg concentrations, from 2×1018 cm-3 to 5×1019 cm-3, coincides with increasing density of small, 3-10 nm-sized, SFs. Emission lines ascribed to SFs are observed in CL in all the studied samples. The observed SF-related luminescence can be explained by a model where Mg atoms interacting with the nearby SF changes the confinement for holes and leads to a pronounced defect-related luminescence. Non-polar m-plane homoepitaxial GaN layers with Mg concentration of 2×1018 cm-3 and 3×1019 cm-3 exhibits high density of basal SFs as well as a number of prismatic SFs. Instead of normally observed in nonpolar GaN SF-related broad lines several sharp lines are detected in the 3.36-3.42 eV region. Their relation to donor-acceptor pair recombination (DAP) was dismissed by calculating the DAP energies and fitting with the measured spectra. The sharp lines are tentatively explained by some impurities bound to point defects or SFs. The origin of two Mg related acceptor bound exciton (ABE) peaks in the emission spectra is also proposed: narrower ABE1 peak at 3.466 eV is identified as coming from a substitutional Mg atom. Broader emission at 3.454 eV is deemed to be coming from a Mg acceptor atom perturbed by a nearby SF. Additionally, Mg cluster formation in the highest doped sample ([Mg] = 1×1020 cm-3) was revealed by APT.

Simultaneous doping by Si and O was studied for HVPE grown bulk GaN. Doping with O concentration from 1017 cm-3 leads to a decrease in the Si concentration to less than 1016 cm-3. Si incorporation is believed to be suppressed by the competing Ga-vacancy-O incorporation process. Bandgap narrowing by 6 meV due to high doping was observed. Donor bound exciton (DBE) lifetime was obtained from TPRL experimental data and it is found to decrease with increasing doping. In non-polar m-plane homoepitaxial GaN Si doping influences the SF-related luminescence. At moderate Si concentrations excitons are bound to

(4)

the impurity atoms or impurity-SF complex. Proximity of impurity atoms changes the potential for SF creating localization for charge carriers resulting in SF-related emission. At dopant concentrations higher than the Mott limit screening destroys the carrier interaction and, thus, the exciton localization at impurity-SF complex.

Finally, C-doped HVPE grown bulk GaN layers were studied by TEM, CL, and TRPL. Enhanced yellow line (YL) luminescence was observed with increasing C doping. Stability of YL in a wide temperature range (5-300 K) confirms that YL is due to a deep defect, likely CN-ON complex. Low-temperature CL mapping reveals a pit-like structure with

different luminescence properties in different areas. DBE emission dominates in CL spectra within the pits while in pit-free areas, in contrast, two ABE lines typical for Mg-doped GaN are observed.

(5)

P

OPULÄRVETENSKAPLIG SAMMANFATTNING

Ungefär 20% av all elektricitet som produceras i världen används till belysning. Det betyder att vi kommer att spara mycket energi om vi ersätter nuvarande lågeffektiva glödlampor med starka vita lysdioder (LED), dessutom kommer utsläpp av koldioxid på planeten att minskas. Detta förklarar orsaken till den enorma utvecklingen av LED-industri som har skett under de sista 20 åren.

En lysdiod är en elektronisk komponent som avger ljus när en elektrisk ström leds i framåtriktningen. Den enklaste typen av diod består av två skikt av halvledarmaterial. Halvledarmaterial leder elektrisk ström dåligt men genom att dopa materialet med små koncentrationer av så kallade dopämnen kan man påverka ledningsförmågan. Dopatomerna kan ge extra elektroner till halvledarmaterialet eller binda elektroner till sig. När en elektron binds, skapas ett så kallat hål (avsaknad av elektron) som också kan leda ström. Materialet som har ett överskott av elektroner kallas för n-typ material och den andra, som har ett underskott av elektroner, – p-typ material. Skiktet där ett p-dopade material övergår till ett n-dopade material kallas för en pn-övergång. Där kan elektroner och hål möta varandra, annihilera och i denna process utstrålas ljus. Genom att använda sig av olika typer av halvledarmaterial kan man skapa lysdioder som lyser i olika färger. Man har upptäckt att man kan tillverka högeffektiva blåa lysdioder i galliumnitrid (GaN) som kan användas som belysningskälla. Genom att låta det blåa ljuset excitera ett fosforskikt kan man omvandla det till vitt ljus. Vita lysdioder baserade på GaN finns redan på marknaden men deras effektivitet och prestanda kan fortfarande betydligt förbättras. Med en förbättrad galliumnitridteknologi kan vi förvänta oss en längre livstid och högreverkningsgrad hos lysdioderna.

Denna avhandling fokuserar på karakteriseringen av tunna och tjocka skikt av GaN som är viktiga för framställning av energisnåla vita lysdioder. Syftet med den forskningen var att få bättre förståelse hur dopning påverkar optiska och strukturella egenskaper hos GaN. Studierna har utförts med hjälp av flera mikroskopiska tekniker, som till exempel transmissionselektronmikroskopi och atomsondtomografi med nära nog atomär upplösning samt med optisk spektroskopi i form av fotoluminescens och katodoluminescens.

Dopning av GaN-skikten med magnesium, kol, kisel och syre har utforskats här. Proverna som har studerats var av både polär och icke-polär kristallorientering. Den sistnämnda orienteringen är viktig för tillverkning av LED-baserad på icke-polär GaN med bättre verkningsgrad. I det arbetet har jag studerat bland annat material som odlas med olika gasfasmetoder på substrat av GaN självt. Detta ledde till en framgångsrik framställning av GaN-skikt med en överlägsen strukturell kvalitet med färre defekter och lägre mekaniska spänningar i materialet.

(6)

(7)

P

REFACE

This Doctoral Thesis is a result of four and a half years’ work during my Ph. D. studies in the Thin Film Physics Group at Linköping University. The work is partially based on my Licentiate Thesis (Licentiate Thesis No. 1520, Linköping Studies in Science and Technology: The Effect of Mg Doping on Optical and Structural Properties of GaN). The project was financed by the Swedish Energy Agency and the Swedish Research Council (VR). The results are presented in seven included papers preceded by the introduction.

(8)

(9)

I

NCLUDED

P

APERS

1. Luminescence related to high density of Mg-induced stacking faults in homoepitaxially grown GaN

S. Khromov, C. G. Hemmingsson, H. Amano, B. Monemar, L. Hultman, and G. Pozina

Physical Review B 84, 075324 (2011)

I contributed to planning, performed large part of characterization (TEM, CL), and wrote the first version of the manuscript.

2. Optical and structural studies of homoepitaxially grown m-plane GaN

S. Khromov, B. Monemar, V. Avrutin, Xing Li, H.Morkoc, L. Hultman, and G. Pozina

Applied Physics Letters 100, 172108 (2012)

I contributed to planning, performed large part of characterization (TEM, CL), and wrote the first version of the manuscript.

3. Luminescence of Acceptors in Mg-Doped GaN

B. Monemar, S. Khromov, G. Pozina, P. Paskov, P. Bergman, C. Hemmingsson, L. Hultman, H. Amano, V. Avrutin, Xing Li, and H. Morkoç

Japanese Journal of Applied Physics 52, 08JJ03 (2013)

I took part in planning, characterization, and discussion of the manuscript.

4. Atom probe tomography study of Mg doped GaN layers

S. Khromov, D. Gregorius, R. Schiller, M. Wahl, M. Kopnarski, H. Amano, B. Monemar, L.Hultman, and G. Pozina

Manuscript in final preparation

I planned the study, did TEM and CL, took part in APT characterization, and wrote the paper.

5. Effect of silicon and oxygen doping on donor bound excitons in bulk GaN

G. Pozina, S. Khromov, C. Hemmingsson, L. Hultman, and B. Monemar

Physical review B 84, 165213 (2011)

I took part in planning, performed TEM characterization, contributed to discussion and to the writing of the paper.

(10)

6. Correlation between Si doping and stacking fault related luminescence in homoepitaxial m-plane GaN

S. Khromov, B. Monemar, V. Avrutin,Fan Zhang, H. Morkoç, L.Hultman, and G.

Pozina

Applied Physics Letters 103, 192101 (2013)

I contributed to planning, performed large part of characterization (TEM, CL), and took part in writing of the manuscript.

7. Effect of C-doping on near-band gap luminescence in bulk GaN substrates grown by halide vapor phase epitaxy

S. Khromov, C. Hemmingsson, B. Monemar, L. Hultman, and G. Pozina

Submitted to Physical Review B

I contributed to planning, performed characterization except for PL, took part in discussion, and wrote the first version of the manuscript.

(11)

R

ELATED BUT NOT INCLUDED PAPER

8. Growth of GaN nanotubes by halide vapor phase epitaxy

Carl Hemmingsson, Galia Pozina, Sergey Khromov, and Bo Monemar Nanotechnology 22 (2011) 085602

(12)

(13)

A

CKNOWLEDGEMENTS

I would like to express my gratitude to people who have helped and supported me during my PhD:

My first supervisor Docent Galia Pozina, for giving me the opportunity to do a PhD in Linköping, for her guidance and support in life and research, huge help with the experimental work and in writing the papers.

My second supervisor Professor Lars Hultman, for always providing good ideas on how to improve my papers and the thesis.

Professor Bo Monemar, for sharing some of his enormous knowledge of

III-nitrides and helping with the interpretation of the experimental data.

Docent Carl Hemmingsson for sharing some of his experience of CVD growth

and providing the samples.

Professors Hadis Morkoc and Hiroshi Amano and the members of their

groups, for providing the samples.

Colleagues in Germany at IFOS, for providing help with the APT measurments.

Thomas, Inger, Kirstin and Camilla, for technical and administrative help.

All the colleagues at Thin Film, Plasma, and Nanostructured Materials groups. I have met many wonderful friends during my time in Linköping and would like to thank all of them for making my time here fun and memorable.

A special thanks goes to my family.

(14)

(15)

1. Introduction

With the invention of light emitting diode (LED) technology and demonstration of the laser diode (LD) in 1960s there began a search for materials that would be suitable to make efficient optoelectronic devices based on these effects. Soon it was understood that II-VI and III-V group materials are appropriate for this goal. III-V semiconductor GaAs and its compounds were used for manufacturing of high efficiency red to yellow LEDs and LDs. To produce high brightness full-color displays or to make high brightness white light sources an efficient blue LED was needed. Possible candidates for short-wavelength LEDs were II-VI compounds like ZnSe and the indirect bandgap semiconductor SiC. Short lifetimes of ZnSe-based devices and low efficiencies of SiC-ZnSe-based LEDs, however, severely limited their use [1]. GaN, on the other hand, looked much more promising for production of short wavelength devices.

GaN is a III-V group semiconductor with a relatively large direct energy bandgap of 3.4 eV. Related compound AlN has 6.2 eV bandgap and InN – 0.64 eV (Fig. 1.1). Synthesizing ternary compounds AlGaN, AlInN, and InGaN provides a large flexibility in bandgap engineering: light emitters and detectors working from near infrared to ultraviolet (UV) regions are possible to produce. Multi-junction solar cell approach based on InGaN, for example, offers possibility of using the whole solar energy range with theoretical efficiencies of 60% [2] compared to theoretical efficiency limit for Si based solar cells at ~30% [3].

(18)

Due to its high vapor pressure of 45000 atm. and high melting point temperature of 2500 °C [4] it is practically impossible to grow single crystals of GaN from the liquid phase. Therefore, methods that employ growth from the vapor phase are used to grow GaN, such as molecular-beam epitaxy (MBE), halide vapor phase epitaxy (HVPE) or metal-organic vapor phase epitaxy (MOCVD). The research on GaN started in the early 1970s when the first violet LED based on HVPE-grown GaN was demonstrated by Maruska and Pankove in 1972 [5]. However, further development of GaN-based devices was hindered in the 1970s and 1980s by low crystalline quality and p-type doping issues.

The introduction of a low-temperature buffer layer of AlN on sapphire substrates at the initial growth stages by Amano and Akasaki in 1986 [6] led to a considerable improvement in crystal quality of epitaxial GaN. Later in 1989 the same group solved the p-type doping problem. Magnesium was known to be a good p-p-type dopant, nevertheless it was difficult to obtain sufficient hole concentration due to passivation of Mg atoms by hydrogen, which comes from precursors during the MOCVD growth. Amano and Akasaki proposed low energy electron irradiation as a means to activate charge carriers thus increasing hole concentration to suitable values of ~1016 cm-3 [7]. Later thermal annealing in N2 ambient was

also suggested to activate the Mg acceptors [8]. As a result of such development Nakamura demonstrated in 1993 the first bright blue and green GaN heterostructure-based LEDs with more than 100 times higher efficiency compared to other alternatives [9].

This breakthrough brought about ever increasing research efforts towards more efficient and brighter blue and white LEDs based on GaN and eventually to a revolution in lighting that we are witnessing right now. Besides solid-state lighting and solar cells, GaN-based devices find applications in full-color LED displays and indicators, data storage (Blu-ray technology uses GaN-based LDs), telecommunications (optical fiber networks), high power electronics, water purification and many others.

One of the remaining challenges on the way to produce better devices is the lack of native substrates for GaN. Instead, heteroepitaxial growth of GaN is usually done on sapphire (α-Al2O3) or silicon carbide (6H-SiC), for commercial purposes. Growth on foreign substrates

leads to a high dislocation density due to a substantial difference in lattice constants (Fig. 1.1) and in thermal expansion coefficients. Threading dislocation density in GaN grown on sapphire is ~108 – 109 cm-2 [10] and it is rather surprising that working devices exist with such a high density of defects. Threading dislocations act as nonradiative recombination

(19)

centers [11] and as scattering centers that affect carrier mobility [12], [13]. Other extended defects include stacking faults (SFs) and associated with them partial dislocations. They are typical for nonpolar grown GaN where the density of SFs and partial dislocations is typically 105 cm-1 and 1010 cm-2, respectively [14]. The structural defects have detrimental effects on optoelectronic devices because they increase the current threshold in emitters, cause leakage in the form of dark current in detectors, and lower working lifetimes.

Another challenge is the development of nonpolar grown GaN and related alloys. High piezo and spontaneous polarization fields present in polar (c-plane oriented) III-N materials cause band bending of the quantum well heterostructures in the active layers leading to non-linear optical effects and lowered radiative efficiency. Nonpolar (m-plane, 1100 , and

a-plane, 1120 , oriented) GaN material do not suffer this problem and consequently underwent

a considerable development during the last 15 years. Initially, devices based on the nonpolar III-N layers were mostly grown on foreign substrates, e.g., (100) γ-LiAlO2 or r-plane

sapphire. External quantum efficiency of such devices proved to be much lower compared to early nonpolar devices grown on native m-plane GaN substrates [15]. Development of nonpolar GaN grown on native substrates is therefore highly desirable.

Mg, C, Si, and O are the major elements that are used for doping in GaN. Mg is the only working p-type dopant successfully used in III-N optical device fabrication. C was found to be useful for producing highly resistive semi-insulating back-barrier layers for high electron mobility transistors (HEMT). Si is the main n-type doping element and O is mainly found as a residual n-type dopant. Better understanding of dopants’ optical signatures, of, in particular, acceptor bound exciton (ABE) lines in case of Mg, donor bound excitons (DBE) in case of Si and O, and emissions related to C would provide ways for improvement of the GaN material, and consequently more efficient devices.

This thesis is focused on the optical and structural properties of bulk, polar and nonpolar homoepitaxial GaN doped with Mg, C, Si or O. The appended papers with research results are preceeded by several chapters giving an introduction to the field. Chapter 2 introduces GaN growth techniques, in Chapter 3 properties of GaN are given, Chapter 4 treats optical transitions dominating GaN optical spectra, in Chapter 5 nature of extended defects is discussed, characterization techniques used to obtain the results are described in Chapter 6 and, finally, and in Chapter 7 a summary of the papers is given.

(20)

2. Growth of GaN

More traditional semiconductors like silicon or gallium arsenide are typically grown from liquid phase by Czochralski or Bridgman methods. GaN, on the other hand, when heated to atmospheric pressure dissociates rather than melts, thus growth from liquid phase at reasonable conditions is impossible. Table 2.1 shows melting temperatures and pressures for several of the most typical semiconductors [16]. For GaN, melting temperature is a theoretically calculated value [4] and melting pressure is an extrapolated value from the experimental data [17]. It can be seen that these values are much bigger compared to common semiconductor materials and rather close to the values of diamond synthesis. Therefore, vapor phase deposition techniques are used for GaN growth.

Table 2.1. Melting conditions of semiconductors [16].

Crystal TM, oC pM, atm. Si 1400 <1 GaAs 1250 15 GaP 1465 30 GaN 2500 45000 Diamond (synthesis) 1600 60000

There exist three major growth techniques for GaN growth: molecular beam epitaxy (MBE), HVPE, and MOCVD. MBE is a non-equilibrium vapor phase epitaxy growth method, where solid or gas source elements are heated in evaporators, to produce beams of atoms impinging on the substrate, which is rotated and heated to appropriate temperature. Deposition occurs under ultra-high vacuum (10-9 Torr) conditions, which is the costly drawback of this method. The advantages are the lower, compared to other methods, growth temperatures of 600-800 °C and possibility to grow very thin layers with fine control of

(21)

composition. Precision of this growth technique is based on low deposition rates and in-situ characterization capabilities. For example, reflection high energy electron diffraction (RHEED) detector allows to control the film thickness up to one atomic layer. Nevertheless this method remains more of a research tool due to quite low growth rates (typically less than 1 µm/h [18]) and rather sophisticated equipment.

Thin and thick films of GaN studied in this thesis were grown by chemical vapor deposition techniques – HVPE and MOCVD – so they will be described in more detail.

2.1 NUCLEATION AND GROWTH

During growth from gas-phase on heterogeneous substrates three different growth modes can take place (Fig. 2.1): (1) island (Volmer-Weber) growth occurs when the deposited atoms are more strongly bound to each other than to the substrate; (2) layer-by-layer (Frank-Van der

Merwe) growth mode occurs when stronger adatom-substrate bonding leads to planar,

layer-by-layer deposition; (3) Stranski-Krastanov mode is a combination of the previous two. The latter starts with several complete monolayers and continues with 3D island growth as layer-by-layer growth becomes energetically unfavorable. When growing semiconductor materials for electronics and optoelectronics, such as GaN, it is a requirement to obtain single crystal

epitaxial films, therefore growth parameters must be adjusted so that layer-by-layer growth

takes place.

(22)

2.2 EPITAXY

Epitaxy is a combination of the two Greek words, επι (epi, meaning “above, upon”) and ταξιζ (taxis, meaning “arrangement, order”). It means the formation of a single-crystal film on top of a substrate material, where crystallographic properties of the grown film are largely influenced by the host material. Two types of epitaxy can be distinguished: homoepitaxy and heteroepitaxy. If the substrate and the deposited film are of the same material then this is called homoepitaxy. The most typical example of homoepitaxy is Si on Si. Higher-quality films, lower defect density and doping control are known advantages of homoepitaxy. Rapid development of Si-based early bipolar transistors and integrated circuits was a consequence of adopting epitaxial methods [19].

Fig 2.2. Three possible cases of heteroepitaxy: matched, strained, and relaxed [19].

If the deposited material is different (but normally of the same crystal structure) from that of the substrate this is the case of heteroepitaxy. When the lattice parameters of the film and the host material are the same, the sort of heteroepitaxy is called to be matched between lattice planes (see Fig. 2.2, a). In most cases lattice parameters of the two materials are different, therefore in heteroepitaxy interfacial bond straining arises between the film and the substrate (Fig. 2.2, b). The film grows in a strained state up to a certain, so called critical thickness, after which the accumulated strain is released either elastically by 3D island formation -

(23)

Stranski-Krastanov growth - or plastically by formation of misfit dislocations - Frank-Van der Merwe growth (Fig. 2.2, c) [20].

Due to lack of native substrates mentioned above, GaN is still predominantly grown heteroepitaxially on SiC (3.5% lattice mismatch) or sapphire (16.09% lattice mismatch). Table 2.2 gives a comparison of GaN lattice and bandgap with those of potential substrate materials. Large lattice mismatch and difference in thermal expansion coefficients between the film and the substrate materials results in high density of misfit dislocations, which lie in the film/substrate interface plane. A different type of dislocations that is reported to be detrimental for electrical and optical propreties of GaN devices is threading dislocations (TDs) [21]. TDs propagate perpendicular to the film/substrate interface. Their origin is, however, controversial, as discussed in Chapter 4.

Table 2.2. Lattice properties and bandgaps of GaN and potential substrate materials [22]. Material (crystal symmetry) Lattice constants, Å Plane with closest match to (0001) GaN Effective a lattice constant, Å Lattice mismatch with GaN (%) (aGaN – asub)/asub

Bandgap, eV GaN (hexagonal) a = 3.1891; c = 5.1855 (0001) 3.1891 0 3.44 Al2O3 (trigonal) a = 4.758; c = 12.991 (0001), rotated 30º 2.747 16.09 >8.5 4H-SiC (hexagonal) a = 3.073; c = 10.053 (0001) 3.073 3.77 3.2 6H-SiC (hexagonal) a = 3.081; c = 15.117 (0001) 3.081 3.51 2.86

As was mentioned in Chapter 1, the introduction of low temperature AlN or GaN buffer layers lead to the dramatic improvement in GaN epitaxial films’ quality. Adoption of low temperature buffer layer techniques allowed 2D Frank-Van der Merwe growth mode to be obtained. At the same time development and industrial application of bulk GaN layers that can be used as native substrates promises even higher light output due to lower number of intrinsic defects, e.g. TDs, that act as non-radiative centers. By using native substrates TDs density, for example, can be lowered to ~106 cm-2 compared to ~109 cm-2 for growth on foreign substrates [23].

(24)

2.3 BASICS OF A CHEMICAL VAPOR DEPOSITION (CVD)

PROCESS

Unlike in physical vapor deposition (PVD) methods where atom species are just deposited on the substrate, in CVD atoms and molecules that are transported to the substrate undergo chemical reactions and thereby form the film. HVPE and MOCVD techniques can be differentiated based on which Ga precursors are used – organic or inorganic. CVD methods allow higher purification of precursors compared, for example, to solid sources used in PVD techniques, which is extremely important due to the necessity of precise dopant control as dopants define electrical and optical properties of semiconductor material.

CVD processes typically includes the following steps for the film growth to occur (Fig. 2.3):

1. Transport of precursors to the reaction zone

2. Chemical reactions of precursors to produce reactive species and by-products 3. Diffusion of the reactants and their products to the substrate surface

4. Physical and chemical adsorption of the reactants on the crystal surface 5. Diffusion and incorporation of the species

6. Desorption and transport of the reaction by-products away from the reaction chamber.

Part of this thesis is focused on HVPE grown bulk GaN (Papers 5 and 7), while another section deals with polar and non-polar GaN films grown by MOCVD homoepitaxially on HVPE substrates (Papers 1-4 and 6).

(25)

Fig. 2.3. Gas transport and reactions during a typical CVD process [24].

2.4 HALIDE VAPOR PHASE EPITAXY

HVPE dominated as a growth method during the early stages of GaN research [25]. Due to difficulties in obtaining p-type conductivity (by doping with Zn or Mg) and poor crystallinity by HVPE attention shifted to MOCVD technique in the 1980s. In recent years HVPE method gained popularity again for the growth of thick GaN layers for substrate applications. This is driven by the development of blue LEDs for high density storage, high brightness LEDs and power devices that need a native substrate to fully realize the potential of the III-nitride materials system [26]. Compared to other III-V material systems progress in bulk GaN lies significantly behind the epitaxial film development. The cost of GaN commercially available substrates is around 100 $/cm2 compared to less than 1 $/cm2 for GaAs [27].

The reasons HVPE is considered advantageous for substrate growth are its inherent high growth rates (up to 300 µm/h), relatively easy and well understood chemistry, possibility of easily scaling up the process and low cost. Films grown with HVPE are also virtually free from carbon due to the absence of carbon species in the growth process (compared to MOCVD). Challenges faced by HVPE method are that the films can be of non-uniform quality, they can exhibit domain structure as well as bowing and cracking due to lattice mismatch and thermal coefficient difference between sapphire and GaN.

(26)

Thick HVPE GaN films for Papers 1, 5, and 7 were grown by Carl Hemmingsson at LiU in a vertical hot wall HVPE reactor. It is schematically depicted in Fig. 2.4. The chamber is made of quartz and is heated resistively and by RF induction. There are two temperature zones: the metallic source zone, which is heated resistively, is kept at the temperature of ~800-900 °C, whereas the growth zone, which is heated by RF induction, is at ~1000-1100 °C.

Fig. 2.4. Schematic drawing of a HVPE GaN growth reactor [28].

Hydrogen chloride is let into the quartz tube with molten Ga where reaction producing gaseous gallium monochloride occurs:

2Ga + 2HCl → 2GaCl + H2 (2.1)

The efficiency of this process is very high, above 95% [29]. In a separate tube NH3,

the group V source, is flowed. Subsequently ammonia and gallium chloride undergo the following reaction on the surface of the substrate:

(27)

The growth process is conducted under ammonia rich conditions, normally NH3 to HCl flow

ratio is 25-30.

Fig. 2.5. Light microscopy image of C-doped bulk (> 2 mm) GaN samples from

Paper 7, front GaN side (a) and back sapphire side (b).

To prevent parasitic growth in the gas inlet ammonia and halide (GaCl) should not be mixed before they reach the substrate. Light molecule gases such as H2 or mixtures of N2 and

H2 are used as carrier gases to obtain laminar flow. The deposition rate is very high compared

to other methods, ~100-300 µm. The process is operated at atmospheric pressure, at near equilibrium conditions and is mass transport limited by the flow of halide precursor.

For C-doping in Paper 7 acetylene gas (C2H2) diluted in N2 was used. In Paper 5

O-doping was achieved by flowing O2 mixed with N2 and Si-doping by flowing silane (SiH4)

mixed with H2. A light microscopy image of C-doped bulk (> 2 mm) GaN samples from

Paper 7, front GaN side and back sapphire side, are demonstrated in Fig. 2.5 (a) and (b). A

combination of HVPE and MOCVD techniques can yield higher quality GaN substrates. Therefore, growth of thick HVPE layers studied in this thesis was always done on a thin (2-3

(28)

2.5 METAL-ORGANIC CHEMICAL VAPOR DEPOSITION

The first time MOCVD was used to grow III-V materials was the growth of GaAs by Manasevit in 1968 [30]. Application of MOCVD to the growth of GaN and AlN was demonstrated by Manasevit et al in 1971 [31], however, due to low purity of precursors and non-optimized processes no semiconductor-quality GaN was obtained. Real breakthrough with the MOCVD growth of GaN was achieved in 1986 by Amano and Akasaki [6] when, now widely used, “two step method” technique was implemented. During “two step method” growth a thin (several tens of nm), low temperature heteroepitaxial GaN or AlN buffer layer is introduced on sapphire and, only upon that, a high temperature GaN film is grown. Low temperature buffer layers play a role as nucleation layer, which absorbs strain formed during heteroepitaxila growth. Solution of the p-type problem [7] and demonstration of the first blue LED [32] contributed to the success of MOCVD. Nowadays it is by far the most commercially used method for III-nitride growth.

In MOCVD, in contrast to HVPE, instead of inorganic species metal-organic precursors are flown in the reactor as sources of Ga. Specifically trimethylgallium ((CH3)3Ga)

is used as Ga, and ammonia – as nitrogen precursors. For Mg doping, organic compounds are used too, specifically for samples in Paper 1-4 bis-cyclopentadienylmagnesium (Cp2Mg) was employed. Alternatively, silane (SiH4) was used as a source for Si (n-type) doping of GaN (Paper 6). Organic species and ammonia are transported in separate pipes to the quartz chamber in order to prevent premature reacting. Lighter carrier gases (H2 or N2) are let

through organic compound liquids thus transporting the precursors to the reactor. These compounds are thermally dissociated over the surface of ~1000 °C hot substrate and react in a sequence of complex reactions forming GaN. In hot wall MOCVD the whole reactor is heated by RF coils, whereas in the cold-wall process only the susceptor, which is usually made of graphite, is heated. The substrate may be rotated to obtain uniform films. Before the growth is started nitridation of the substrate is usually done to reduce lattice mismatch due to the nitrides formed on the substrate surface (SiN on SiC or AlN on Al2O3). The usual practice,

and how it was done for the growth of the samples studied in this thesis, is to start epitaxial growth on sapphire substrates with a low-temperature (~600 °C) GaN buffer layer (“two step method”).

(29)

Ease of scalability, high uniformity, and quality of the films as well as a relatively high growth rates (~1 µm/h) are making this technique the method of choice for commercial purposes. Undoped 0.5-µm-thick buffer layer and 1-µm-thick Mg-doped epitaxial layer for samples in Paper 1, 4 were grown by MOCVD. In Paper 2 400 nm Mg-doped layer, starting with an undoped 0.6 µm GaN layer, and in Paper 6 1-µm-thick Si doped m-plane GaN layers were grown by this method as well.

Fig. 2.6. Schematic drawing of a MOCVD reactor that was used for growth of GaN epilayers. TMG stands for trimethylgallium, TMA – for trimethylalluminium and Cp2Mg – for bis-cyclopentadienylmagnesium.

(30)

3. Properties of GaN

Atomic bonds in III-V semiconductors are covalent with some degree of ionicity. Compared to other group V-elements, nitrogen has a smaller atomic radius and a lower Coulomb screening due to a smaller number of electrons, meaning that the electrons are tightly bound. This makes N a very electronegative element bringing a much higher degree of ionicity to the III-N bond. According to Pauling ionicity criteria GaN has 0.387 ionicity compared to 0.039 for GaAs. This high ionicity leads to a stronger bond strength and a shorter bond length, which in turn, explain some of the macroscopic physical properties of GaN, such as significant hardness and chemical stability. Other superior material properties of GaN are high electron mobility, high thermal conductivity, high critical field and direct wide bandgap. These properties are presented and compared to the properties of other major semiconductors in Table 3.1.

Table 3.1. Important properties of GaN at 300 K [33]–[35], [36].

Material Bandgaps, eV Lattice constants, Å Mobility, cm2·V-1·s -1 Thermal conductivity, W·cm-1 ºC-1 Breakdown field, V·cm-1 GaN 3.4 direct a = 3.189 c = 5.186 electrons 1000 holes 200 1.5 > 5×106 Si 1.1 indirect 5.431 electrons 1400 holes 450 1.5 3×105 GaAs 1.4 direct 5.653 electrons 8500 holes 400 0.5 4×105 6H-SiC 2.9 indirect a = 3.073 c = 10.053 electrons 600 holes 40 5 4×106

Direct bandgap is an extremely important property for a material aimed to be used in optoelectronics. Direct bandgap means that the minimum of the conduction band and the maximum of the valence band are aligned in the reciprocal space. Indirect bandgap, on the other hand, means that they are separated by some ∆k value. Schematic examples of direct

(31)

and indirect bandgaps are presented in Fig. 3.1. During the operation of an optoelectronic device holes and electrons are injected into the p-n junction. To efficiently recombine and produce photons they should be at the same position in the reciprocal space. In indirect bandgap materials a hole and an electron are separated and recombination can occur only by emitting or absorbing a phonon that would compensate ∆k difference. This process is, however, orders of magnitude less probable than the recombination process in a direct bandgap material.

Therefore, mainly direct bandgap semiconductors are suitable for the production of highly efficient optoelectronic devices. Even though 6H-SiC has similar bandgap value to GaN and advantageous properties such as thermal stability and high breakdown field, indirect bandgap was one of the reasons that it was largely discarded in favor of GaN for the production of blue LEDs.

Fig. 3.1. Schematic pictures of (a) direct and (b) indirect bandgaps in semiconductors.

High breakdown field, a consequence of the large energy bandgap, allows higher voltages to be applied to GaN based devices. High charge carrier mobilities offer possibilities of high frequency applications to be realized. High thermal conductivity is beneficial in case of high working temperatures and where fast and efficient heat dissipation is required. These are the reasons to use GaN for making next generation high power/high frequency microwave devices.

(32)

3.1 CRYSTALLOGRAPHIC PROPERTIES

GaN atomic bonds are sp3-hybridized which dictates tetrahedral coordination of the atoms, i.e. each Ga atom is surrounded by four N atoms and vice versa. GaN has two polytypes where atoms are tetrahedrally coordinated: a cubic, zinc blende phase (space group

F43m) and a wurtzite, hexagonal phase (space group P63m). Cubic form of GaN is only

metastable and difficult to grow, most of the research is therefore focused on wurtzite GaN. All the results in this thesis are also solely obtained from wurtzite GaN. The wurtzite crystal structure can be imagined as two interpenetrating hexagonal closed packed (hcp) lattices, consisting of the corresponding element atoms, Fig. 3.2.

Fig. 3.2. Unit cell of wurtzite GaN. The four basis atoms are highlighted.

The basis (highlighted in Fig. 3.2) consists of four atoms, two of each type, their coordinates are the following: Ga atom (0, 0, 0), (1/3, 2/3, 1/2) and N atom (0, 0, u), (1/3, 2/3, 1/2 + u), where u is the distance between Ga and N atoms in [0001] direction and it is equal to 3/8 of the unit cell constant c. The value 3/8·c is only valid for an ideal wurtzite structure, however, and it varies slightly with the degree of non-ideality of the crystal. For GaN, the value of u is 0.376 in unstrained material. There are two different positions for atoms in the wurtzite structure, therefore the stacking sequence can be written as AaBbAa, where capital letters denote Ga atoms and small ones – N atoms. It is interesting to compare stacking sequences of zinc blende and wurtzite structures, Fig. 3.3, because it can help understand how

(33)

SFs are formed, for example. Zinc blende structure, in difference to wurtzite, consists of three alternating bilayers of atoms: AaBbCc.

Figure 3.3. Stacking sequences of wurtzite and zinc blende GaN.

3.2 P

OLARITY AND

P

OLARIZATION

GaN lacks inversion symmetry in the [0001] direction in wurtzite. This means that

[

0001

]

and 0001_ _{ directions are not equivalent: if we flip the crystal in this direction Ga} atoms will end up in N atom’s places and N atoms – in Ga atom’s places. Thus GaN can be grown in two different orientations in [0001] direction: Ga-faced GaN with [0001] orientation and N-faced GaN with 0001_ _{ orientation. It is important to point out here that Ga-face does} not mean that the surface is terminated by Ga atoms (Fig. 3.4). For example, N-face surface can be terminated by Ga atoms. Instead, polarity is defined by the direction of the Ga-N bond. In this Thesis, I have been concerned primarily with Ga-face GaN layers.

(34)

Fig. 3.4. The two different orientations and spontaneous polarization vector in GaN.

High ionicity of the Ga-N bond means that the bonds act like microscopic dipoles. This, in combination with lack of inversion symmetry results in the presence of macroscopic spontaneous polarization in [0001] direction. The word “spontaneous” means that it is present in the absence of strain.

There are 4 bonds pointing in different directions in a GaN tetrahedron. Therefore, the polarization contributed by the vertical bond is counteracted by the polarization contributed by the other three bonds, minimizing the overall spontaneous polarization effect. The degree of minimization is dependent on the angle between the bonds, which in turn, depends on the degree of non-ideality of the crystal, i.e. on deviation of u value from 3/8· c and c/a from c0/a0.

The degrees of non-ideality and its influence on the spontaneous polarization in GaN, AlN, and InN are summarized in Table 3.2 for comparison.

Table 3.2. Lattice parameters deviation from the ideal wurtzite structure and its influence on the spontaneous polarization in III-nitrides [37].

Material ideal GaN AlN InN

u0 0.375 0.376 0.380 0.377

c0/a0 1.633 1.6336 1.6190 1.6270

(35)

Presence of external stress due to lattice mismatch in the films grown on foreign substrates, or in heterostructures, results in additional, piezoelectric contribution to the polarization. The total polarization, therefore, can be presented as the sum of the two components:

P =Ptot spon+Ppiezo (3.1)

The direction of piezoelectric polarization is dependent on the type of stress that is present. In AlGaN, which has a smaller lattice constant, when it is grown on GaN, the resulting stress is tensile. In InGaN grown on GaN, conversely, the strain is compressive due to a larger than in GaN lattice constant. The resulting piezoelectric field in these materials has different signs: it is parallel with spontaneous polarization in AlGaN and antiparallel with it in InGaN. The overall polarization effect in InGaN is therefore smaller compared to AlGaN.

The polarization fields have very important implications for GaN LED devices. These devices are based on quantum well (QW) heterostructures, where thin layers of material with a narrower bangap are embedded in a wider bandgap material matrix, for example InGaN QW in GaN matrix or GaN QW in AlGaN matrix. QW structures confine the charge carriers and dramatically increase the probability of their radiative recombination. The presence of spontaneous and piezoelectric polarization in QW leads to bending of the bandgap therefore spatially separating the charge carriers (Fig. 3.5).

(36)

Spatial separation of the charge carriers results in a smaller overlap of the wave functions and consequently in lower recombination probability. Additionally, it red-shifts the output emission, i.e. the emitted photons are of lower energy [38]. This phenomena is called quantum confined Stark effect and is detrimental for optoelectronic devices as it lowers the overall efficiency.

Besides that, tilting of the band edges of the active layer makes the effective width of QWs smaller, which results in higher density of charge carriers and increases probability of non-radiative Auger recombination. Band bending also lowers an effective barrier height in QWs leading to carrier leakage with increasing bias voltage. These two phenomena are believed to cause the so-called “efficiency droop” effect in GaN based light emitters when external quantum efficiency of devices drops with increasing current [39], [40].

While unwanted in optoelectronic devices, built-in electric fields were found to be beneficial for high electron mobility transistors (HEMT), which employ AlGaN/GaN heterojunctions. Polarization field, which is higher in AlGaN than in GaN creates a sheet charge at the AlGaN/GaN interface. This charge is compensated by the mobile electrons from

n-type doped AlGaN that diffuse into QW at the interface and end up in semi-insulating GaN

forming the so-called 2-dimensional electron gas (due to confinement in one direction). Ionized impurity atoms from n-type dopants in AlGaN and electrons in the QW are therefore spatially separated. This eliminates the impurity scattering bringing much higher carrier mobilities.

3.3 NONPOLAR GROWTH DIRECTIONS

To solve the problem of high polarization fields present in c-plane GaN-based heterostructures, growth on nonpolar planes was proposed [41]. Due to inversion symmetry, polarization fields are compensated in these directions, hence the name nonpolar. There are two nonpolar sets of planes: a-planes,

{ }

1120 , and m-planes,

{ }

1010 , Fig. 3.6.

(37)

Figure 3.6. Polar (c-) and nonpolar (m- and a-) planes in GaN.

A seminal paper was published by Waltereit et al. [42] in 2000, where nonpolar m-plane heterostructures of GaN/AlGaN grown on tetragonal LiAlO2 were reported. After this

publication interest to nonpolar GaN significantly increased and working devices based on m-plane GaN were produced recently [43]–[45].

Heteroepitaxially grown nonpolar GaN layers are known to exhibit poor surface morphology, with striations and pits [46], [47]. Higher concentration of extended defects compared to c-plane films, especially SFs, is also a problem of nonpolar GaN. Nonpolar films with improved morphology and microstructure can be obtained if they are grown on native GaN substrates [48]. At the moment nonpolar GaN substrates are prepared simply by cutting the c-plane oriented thick HVPE grown GaN substrates along nonpolar direction. In this way small, typically 5×10 mm, nonpolar substrates can be obtained. Due to the small size their availability is still limited and the price is high. m-plane homoepitaxially grown GaN layers doped with Mg and Si were investigated in Papers 2 and 6 of this thesis, respectively.

3.4 DOPANTS IN GAN

There are two types of dopants in semiconductors: donors and acceptors. Donors are atoms of a chemical element that have more electrons in the outer electron shell than atoms of the host material, whereas acceptors are atoms of an element with lower number of electrons in the outer shell. Thus, donor atoms can provide an excessive electron to the material and acceptors can provide an excessive hole. One of the most important requirements for donor or acceptor

(38)

atoms is that they should introduce sufficiently shallow energy levels in the bandgap to be able to provide a reasonable number of electrons and holes into the material at room temperature. Other factors that determine if an element can be used as a dopant are its abundance in nature, cost, or purely technological issues, like solubility or diffusivity.

Due to a very large (3.4 eV) energy bandgap the intrinsic carrier concentration in GaN is very low. To create a p-n junction and consequently a working semiconductor device, as in any other semiconductor material incorporation of dopant elements with reasonably shallow energy levels is needed. It should be noted here that GaN is never free of background impurities, for example Si that can come from the quartz parts of the growth reactors. Therefore, without intentional compensation by other impurities undoped GaN is usually n-type and is often denoted in the literature as unintentionally doped (UID) GaN. Efficient p-type doping, in contrast to n-type doping is more problematic.

Several elements from group II and IV of the Periodic table were tried as p-type dopants in the past. After initial attempts with Zn, Cd, and Be [49] Mg was established as the only working p-type dopant. Mg is a group IIA element, an alkaline earth metal, atomic number 12, one of the most abundant elements on Earth. When it substitutes Ga in GaN lattice it acts as an acceptor and creates quite a deep level, ~230 meV. Mg atoms are known to be passivated by H that is always present as carrier gas in HVPE and MOCVD processes. To break Mg-H complexes thermal annealing or low energy electron-beam irradiation (LEEBI) should be used. Hole concentrations were shown to increase from 2×1015 cm-3 to 3×1018 cm-3 after LEEBI [50]. Due to the deepness of the introduced level high concentrations of Mg (1019 -1020 cm-3) must be incorporated to obtain reasonable hole concentrations. In this thesis Mg doping and its implications for optical spectra and structural properties of both polar and non-polar homoepitaxial GaN are considered in Papers 1-4.

Silicon is a widely used material in electronics, a group IVA element, atomic number 14. As it is situated between period III material Ga and period V material N it can act as both donor and acceptor depending on which atom it substitutes. First principles calculations show however that substitution of Ga is more preferential [51]. Thus upon substitution it creates a shallow donor level ~30 meV below conduction band [52]. Oxygen is a group VA element, it substitutes N and acts as a donor with a shallow level at ~33 meV [52]. Both Si and O can be present even in undoped samples and make it UID. Intentional doping by O and Si in HVPE grown bulk GaN layers was considered in Paper 5 of this

(39)

thesis. Influence of Si doping on optical and structural properties of nonpolar m-plane homoepitaxial GaN was studied in Paper 6.

Carbon lies just above Si in IVA group of the Periodic table and therefore possesses amphoteric properties as a dopant, it can become a donor when it substitutes Ga and an acceptor when it substitutes N. Theoretical calculations have shown, however, that substitution of N is preferred [53]. Due to a small radius it can also take an interstitial place. Initially C was considered as a potential p-type dopant, but all efforts to obtain p-type GaN:C resulted in highly resistive, semi-insulating layers. These layers turned out to be useful in HEMT devices. High resistivity may be attributed to self-compensation of CN with CI [54] or

CGa [53]. There is a controversy whether C creates shallow or deep acceptor level in the

bandgap. A more detailed discussion of substitutional C and its influence on yellow luminescence (YL) line and near bandgap emission (NBE) in HVPE grown bulk GaN are presented in Paper 7.

(40)

4. Optical Transitions in GaN

We can excite electrons from the valence band to the conduction band by means of heat, electromagnetic energy or electron bombardment. To release the excess energy electrons and holes move to their equilibrium states. They can do it non-radiatively, by interacting with phonons, or radiatively, i.e. by emitting photons. Non-radiative recombination is considered unwanted as it produces heat instead of light and lowers quantum efficiency. In case of excitation by light the emitted luminescence is called photoluminescence and in case of excitation by electron beam the emitted light is called cathodoluminescence.

During the radiative recombination excited electron can return directly to the valence band. The emitted photons in this case have the energy equal to Eg. This luminescence is

called band edge luminescence. Another situation is when the excited charge carriers get trapped on the shallow impurity levels. Recombination involving electrons and holes trapped on the impurity levels or annihilation of excitonic complexes bound to impurities provide a way to probe different dopants introduced into the material. Radiative transitions related to impurities are shown in Fig. 4.1. They are discussed in more detail below.

(41)

4.1 DONOR-ACCEPTOR PAIR RECOMBINATION

Donor and acceptor levels are depicted by the short lines inside the bandgap (Fig. 4.1). A direct recombination of excess electron from the donor level with the hole bound to the acceptor level is possible and called donor-acceptor pair (DAP) recombination. Donor and acceptor atoms are neutral before and become charged after the recombination, therefore Coulomb interaction energy is added to the radiative recombination energy:

0 0 2 0 4 DA g D A LO DA e E E E m r ω ω πεε = − − + − ℏ ℏ _{, (4.1)}

where ℏ is Planck’s constant, E_g,E , and _D0 E are bandgap energy, donor activation energy, _A0

and acceptor activation energy, respectively, e is electron charge, εis dielectric constant of the material, ω_DA ,ω_LO are DAP transition frequency and longitudinal optical phonon frequencies, and m is an integer number.

DAP lines in GaN appear at 3.25 – 3.27 eV followed by two LO phonon replicas due to phonon-assisted recombination. DAP emission involves recombination of holes from a shallow acceptor, e.g., Mg with activation energy ~230 meV, and shallow donor, e.g., Si with activation energy ~30 meV or O with activation energy ~33 meV [55]. DAP can only be detected at lower temperatures as it quenches at ~150 K.

4.2 E

XCITONS IN

G

A

N

When an electron is excited to the conduction band leaving a hole in the valence band, three fundamental processes can occur. Firstly, if the electron and the hole get drawn back together by the Coulomb force the electron-hole pair annihilates giving off a photon, which contributes to the band edge luminescence. Secondly, if the electron has high enough energy that it can “escape” from the hole – the electron and the hole become free charge carriers. Then if we separate them, by for example applying an electric field, they give rise to photocurrent – this is how photodetectors and solar cells basically work. The third case is

(42)

intermediate: the electron in the conduction band and the hole in the valence band can be bound together by an electrostatic Coulomb attractive force, forming an electron-hole pair, which is called an exciton. The exciton can move through the crystal and can be characterized by wave vector K. Excitons are unstable and undergo recombination process. The lifetime of free excitons in GaN is of the order of hundreds of picoseconds [56].

Thus, exciton is an electrically neutral quasi-particle consisting of the bound an electron and a hole separated in space. Generally excitons are divided into two kinds: more tightly bound, the so-called Frenkel excitons, with typical binding energy in the range 0.1-1 eV and more weakly bound, Wannier-Mott (WM) excitons, with typical binding energy of ~0.01-0.03 eV. GaN possesses a reasonably high dielectric constant, ε = 8.9, therefore electrostatic field in electron-hole pair is reduced and only weakly bound WM excitons with energies of ~25 meV are present.

Exciton movement can be described by the following Schrödinger equation:

( )

(

)

( )

2 2 2 2 2 0 , , 2 2 4 g e R r R r E E R r M µ πεε r ψ ψ   − ∇ − ∇ − = −     ℏ ℏ , (4.2)

where the first term in Hamiltonian is kinetic energy of exciton translational movement with a center of mass coordinate * * * *

( e e v v) / ( e v)

R= m r +m r m +m , the second term is kinetic energy of rotational movement of electron around hole with a relative radius vector r= −re rh, and the

third term is Coulomb interaction potential; me* and mh* are the effective masses of electron

and hole, * *

e h

M=m +m is the mass of an exciton, * *_{/ (} * *₎

c v c v

m m m m

µ= + is reduced electron-hole mass. Energy E is counted from the top of the energy bandgap Eg.

Translational and rotational movements are independent, therefore the wave function can be presented as follows:

ψ( , )R r =χ( ) ( )Rϕ r (4.3) Then if we divide eq. (4.2) by eq. (4.3) we can get two independent equations:

2

2 _{( )} _{( )}

2M Rχ R Wχ R

(43)

2 2 2 0 ( ) ( ) ( ) 2 r 4 e r r r r ϕ ϕ εϕ µ πεε −ℏ ∇ − = (4.5)

Equation (4.5) is almost the same as for electron movement in the hydrogen atom. Compared to hydrogen, instead of an electron, a quasi-particle with mass µ, is rotating around positively charged center and electrostatic attraction is diminished by value of dielectric constant of the host material ε. The sum of energy eigenvalues W and ε are equal to the energy in (4.2):

W+ = −ε E E_g, (4.6) where 2 2 2 K W M

=ℏ is kinetic energy of the exciton and

4 2 2 2 2 0 8 e h n µ ε ε ε = − is excitonic potential

energy. The dispersion relation for excitons can therefore be expressed from (4.6) as follows:

4 2 2 2 2 2 2 0 8 2 ex g e K E E h n M µ ε ε = − +ℏ (4.7)

The second term in expression (4.7) can be rewritten as

* 2 Ry n , where 4 * 2 2 2 0 8 e Ry h µ ε ε = is

excitonic Rydberg, i.e. excitonic ground state energy at n = 1. Compared to hydrogen, exciton binding energy is weaker by ₂

e m

µ

ε . Excitonic spectra can therefore be imagined as a

hydrogen-like series of emission lines with n = 1, 2, 3, … (Fig. 4.2). The valence band in GaN is split into three subbands due to the crystal-field splitting and spin-orbital coupling [57], consequently excitons involving a hole from each of the subbands can be observed – A, B, and C.

Fig. 4.2. Schematic drawing of energy dispersion curves of GaN at Г point of Brillouin zone.

(44)

Exciton generated through absorption of a photon should have the same impulse value

K = Kph. Photon impulse is very small in the visible area, as ph 0 h K

c

ν

= ≈ , therefore excitonic

impulse is also K ≈ 0. This fact means that excitonic transitions are only possible to the bottom of the excitonic energy zones resulting in characteristic sharpness of excitonic emission lines. At K ≈ 0 where direct transitions occur, excitonic spectra can be described then as follows: * 2 ex g Ry E E n = − (4.8)

By analogy with hydrogen atom excitonic Bohr radius can be written as:

2 * 0 0 0 2 4 e m a a m e πε ε µ = ℏ = (4.9)

Bohr radius is also modified and is larger than in hydrogen by a εm0

µ factor. Typical

excitonic Rydberg energies and Bohr radii in semiconductors are in the range: *

1 meV ≤ Ry ≤200 meV ≪ Eg (4.10)

_{1 nm} * _{50 nm}

lattice

a < ≤ ≤a (4.11) For GaN, these values are: _Ry*=_{25.2 meV} _and _a*≈_{30 Å} _{[58]. Ionized dopants in}

semiconductor contribute charge carriers that at high concentrations can screen, i.e. diminish, the Coulomb interaction between holes and electrons thereby destroying WM excitons. Parameter that describes the charge carrier screening is called the Debye length:

0 2 kT d e N εε = (4.12)

If a*>dthen excitons get destroyed by the Coulomb screening. As it follows from formula (4.12) two factors define the existence of excitons: temperature T and concentration of charge carriers N. Generally excitonic lines can therefore be only observed at lower temperatures and in materials with relatively low charge carrier concentrations. In GaN we can observe excitonic emissioins at room temperature, as Ry*>25 meV.

(45)

4.3 BOUND EXCITONS (BE)

Dopant atoms, incorporated into GaN, have different electronegativity compared to Ga or N therefore they can act as small potential wells, thus, creating a trap either for a hole or en electron. The trapped hole or electron can then attract an electron or a hole respectively and create a BE state. Excitons can be bound to neutral or ionized donor or acceptor atoms or other point defects, e.g., vacancies, or even extended defects, e.g., SFs. Binding of exciton to neutral acceptor and donor atoms and their recombination process are depicted in Fig. 4.3. When BEs recombine they emit energy that can be expressed as follows:

hω = Eg – Eex – EBX , (4.3)

where EBX is the binding energy of the exciton to the donor or acceptor atom. BE lines

therefore appear before the free exciton (FE) lines in the emission spectra. Because BE are more localized than FE their kinetic energy is lower than FE kinetic energy, consequently BE emission lines are sharper then FE lines. BE emission is important for characterization purposes since it can serve as optical signature for defects and impurities.

Fig. 4.3. Schematic drawing of bound exciton electronic structure and the corresponding recombination process.

(46)

DAP, FE, and BE emissions are commonly called near bandgap emission (NBE). A typical spectrum of NBE from c-plane HVPE-grown GaN layer is shown in Fig. 4.4. Besides DAP (at the inset, larger scale), FE lines, acceptor and donor BE we can see longitudinal optical (LO) phonon replicas of the same peaks in the 3.35 – 3.40 eV region. LO phonon interaction induces phonon-assisted exciton emissions accompanying the exciton recombination. The ratio of the exciton peak intensities to the intensities of phonon replicas can serve as a measure of exciton-phonon interaction.

Fig. 4.4. Low-temperature photoluminescence spectrum of nominally undoped HVPE GaN layer grown on sapphire. Courtesy of Dr. G. Pozina.

4.4 YELLOW LINE (YL)

The ubiquitous YL is very often present in GaN emission spectra. It is centered around 2.2 eV and has a broad near Gaussian form. It is detrimental to device efficiency as it competes with the NBE emission that produces the needed UV and violet light. YL was observed in both undoped, i.e. unintentionally n-type doped samples, and in GaN doped with different dopant impurities. Several researchers reported enhancement of the YL due to C doping [59], [60]. Some others reported its relation to Gallium vacancies VGa [61] or to

(47)

controversial. Fig. 4.5 shows a typical spectrum from C-doped HVPE grown bulk GaN with YL and FE emission lines.

(48)

5. Extended Defects in GaN

It was mentioned in the introduction that low availability of cheap and high quality native substrates drives manufacturers to grow GaN on foreign substrates, such as SiC or sapphire. It results in high strain fields due to lattice mismatch and difference of temperature expansion coefficients between the epitaxial film and the substrate, which in turn leads to high density of extended defects, such as threading dislocations (TD) and SFs. Extended defects proved to have limited effect on device performance and working devices exist despite extremely high density of TDs in the range 108 – 109 cm-2 [10] and SFs in the range 105 – 106 cm-2 [64]. However, they do influence device performance, e.g. TDs can act as scattering centers thus lowering carrier mobility [65], and non-radiative recombination centers which lowers quantum efficiency [11]; SFs were linked with higher leakage currents [66]. Therefore, to produce new generations of LEDs, the density of extended defects must be lowered [67] and a lot of research is focused in this direction.

5.1 THREADING DISLOCATIONS

There are normally three types of TDs in c-plane GaN (Fig. 5.1): perfect edge (a-type) dislocations with Burger vector b = 1 1120

3 and line vector J = 0001 , they constitute 40-70% of all dislocations [68]; perfect screw dislocations (c-type) with b = 0001 and J =

0001 , they are the most uncommon type and are 1-2% of all dislocations [65], [69]. The

rest are mixed type dislocations (a+c-type) with b = 1 1123

3 and J about 12° inclined from the [0001] direction [65].

(49)

Figure 5.1. Dislocation types in GaN and their corresponding Burger vectors b and line

directions J. a) Edge dislocation (a-type). b) Screw dislocation (c-type). c) Mixed dislocation

(a+c-type).

There is controversy regarding the origin of TDs. Ning et al. [70] proposed that the rotation of the coalescence islands during the initial growth stages results in TD formation. They suggested that rotation of the islands around [0001] axis leads to their tilt and consequent formation of edge TDs while rotation around axes perpendicular to [0001] leads to their twist and screw TDs formation (Fig. 5.1). These findings were supported by Wu et al. [71]. On the other hand Narayanan et al. [72] in their studies stated that TDs come from the defects at the sapphire/GaN boundary.

Figure 5.2. a) Perfect edge dislocation. b) Perfect screw dislocation.

Different methods are used to lower the TD density, such as epitaxial lateral overgrowth, pendeo epitaxy [73], applying high-temperature AlN buffer layers [67] or transitional metal nitrides interlayers [74]. These techniques allowed to lower the dislocation density by a few orders of magnitude to ~106-107 cm-2.

Doping effects on the structural and optical properties of GaN

Linköping Studies in Science and Technology

Dissertation No. 1554